AI, Innovation and Modernisation - Key Insights from AWS re:Invent 2024

AWS has recently concluded their annual re:Invent conference in Las Vegas and I am excited to share my thoughts on some of the incredible new announcements and updates from the event. I am a Lead Software Engineer at Kainos and have been using AWS technologies for nearly 10 years. This was a pivotal year for AWS re:Invent and features critical announcements that I believe are going to have a significant impact on how industry leaders think about digital transformation, cloud services and AI going forward.

Last year, concerns emerged among industry leaders that AWS was trailing behind competitors such as OpenAI and Microsoft in the AI space. For example, major announcements at re:Invent 2023, such as Amazon Q, Bedrock and Guardrails were still in preview. Furthermore, despite Amazon Q being one of its big announcements, early reviews from users reported that its chatbot app was often slow, inaccurate, and fell short of leading solutions like ChatGPT.

This year, AWS has raised the bar by making AI a first-class priority, with their enhanced services reflecting this commitment. This has been solidified in the introduction of ‘Inference’, their fourth building block alongside Compute, Storage and Databases.

Amazon Nova

Nova Micro, Lite, Pro – general availability, Nova Premier coming Early 2025

Over the week-long conference, AI was at the core of many of AWS’ big announcements. The introduction of Amazon Nova, a new foundational model, is positioned to compete with other leading models such as OpenAI’s GPT or Google’s Gemini. Notably, Nova is AWS’s own model, built from the ground up and it is pitched as being a cost-effective solution, without compromising on performance or quality.

We expect the top selling points of Nova to be:

• With 4 flavours of Nova (Micro, Lite, Pro and Premier), users have the flexibility of picking the right model for the task. For example, Micro is a lightweight model for text-only interactions, while Lite and Pro offer multimodal support. All three of these models are available for use now, while Nova Premier, which specialises in complex reasoning, is expected in early 2025.
• The cost structure for the Nova models stands out significantly from its competitors. For instance, Nova Micro charges approximately $0.14 per million output tokens, making it over 75% cheaper than GPT-4o Mini, which costs $0.60 per million output tokens.
• As an AWS native model, we expect AWS to create an environment which is optimised for Nova, further maximising capabilities, such as performance, latency and updates. Nova Micro can deliver responses with a throughput of 210 tokens per second – 146% faster than GPT-4o Mini’s 85.2 tokens per second.
• The high throughput, cost efficiency and integration with the AWS ecosystem allows Nova to be optimised for RAG and agentic applications, unlocking a tailored and action-orientated approach. For example, Nova can natively and seamlessly integrate with data sources such as DynamoDB or S3, or with workflows using Lambdas and Step Functions.
• Models such as Nova Lite and Pro offer multimodal support for images, audio, video and more. This is something that is evolving rapidly across the industry and makes Nova’s move into this space a logical step forward.

This combination of variety, performance and action holds the potential to unlock significant value across a wide range of use cases. This could be personalised recommendations transforming public sector services, streamlined care plan coordination revolutionising healthcare, or optimised payments management reshaping banking operations. Many organisations grapple with realising the tangible benefits of AI, but this cohesive suite of tools - designed with action at its core - has the power to deliver measurable, transformative value.

AWS has benchmarked its Nova models across various industry-standard metrics against other models to demonstrate its potential, to which it has excelled in many areas making it a serious consideration for modern AI applications. Kainos are working to test these boundaries and give our customers the benefit of our first-hand experience.

Breadth of Model Availability

Amazon Bedrock Marketplace – general availability

AWS has taken the approach that “choice matters” when it comes to AI models. Over 100 models will be available through its Bedrock Marketplace which has the potential to see Gen AI applications built upon more specialised models, e.g. IBM Granite for enterprise workflows, Evolutionary Scale ESM3 for analysing protein structures, NVIDIA Nemotron-4 for multi-language support.

Accuracy and Safety

Bedrock Guardrails – general availability, Automated reasoning – preview

Beyond showcasing the potential of AI models and their capabilities, AWS has enhanced its existing tooling with new features to improve the safety and accuracy of AI applications. Bedrock Guardrails allows developers to implement policies against undesirable content and Automated Reasoning uses mathematical validation checks to reduce hallucinations and improve trustworthiness. It’s worth noting that while AWS's Automated Reasoning checks enhance the accuracy of AI-generated content, it's crucial to recognise that their effectiveness is closely tied to the quality of the inputs and policy definitions set by the user.

SageMaker

SageMaker Unified Studio – preview, SageMaker Lakehouse – general availability

Within the Data & AI space, there is a drive to simplify the solution delivery and speed up value extraction from data. We noted three major areas of interest in updates to SageMaker:

• SageMaker is now being extended to support the development of new foundational models.
• SageMaker Unified Studio is a streamlined development environment combining standalone data and ML tools into a single interface. This was driven by customer demand for a unified and collaborative experience across their data, analytics and ML workflows.
• SageMaker Lakehouse aims to unify data access across a data lake/data warehouse with S3 and Redshift. This enables it to compete with data-centric workflows provided by Databricks & Fabric. The introduction of Apache Iceberg as a first-class citizen enables zero-ETL access for data science/advanced analytics workloads. This also features the ability to time travel on data assets for more complex data. It is worth noting that this Lakehouse update is available immediately without changes to data architecture.

The major updates to SageMaker signify AWS’ attempts to remain a key player in the data and analytics environment. Unified Studio brings SageMaker closer to the more user-friendly and collaborative experience of Microsoft Azure ML Studio. This experience enables robust data governance through native DataZone integration. We can finally translate business processes into policies and requests executed on the platform.

The composition of SageMaker with the latest additions and Lake Formation, DataZone simplifies greatly problems that were challenging before on AWS. Implementation of data lineage becomes configuration of tools instead of custom/costly development
Our team is excited about the ways we can deliver to customers complete, unified Data & AI experiences on AWS with strong data governance foundations.

Through our extensive AWS programme engagement, we were part of the community to gain early access to the beta iteration of SageMaker Unified Studio. Due to our breadth of technical expertise, we were asked to provide our perspective on the tool against existing market solutions to implement it into a customer solution. Before its official launch at re:Invent, we deemed the product not mature enough against the competition, but the consistency and speed with which AWS reiterate means it’s unlikely to be long before it’s a serious contender.

Migrate and Modernise

Amazon Q for Java migrations – general availability
Amazon Q for .NET migrations – preview
Amazon Q for VMWare/Mainframe migrations – preview
GitLab Duo with Amazon Q – preview for GitLab Ultimate subscriptions

AWS’ heritage is in engineering and building for the developer. They understand the developer challenge of legacy technologies across the industry, and how a migration can lead to more performant and cost-effective apps.

They have unleashed that knowledge in the tools that they have just released:

• Amazon Q Developer is a tool to either automate the process of migrations entirely for simple technologies like Java or to provide a migration plan for more complex migrations such as VMWare or Mainframe.
• GitLab Duo – utilising Amazon Q-powered AI agents to perform complex, multi-step tasks such as feature development and technology upgrades (e.g. Java 8 to 21 migrations), all within the standard GitLab UI.

AWS reference their own use case for Java migration. They claim to have used Amazon Q developer to migrate over 10,000 Java 8 applications, which they estimate to have saved over 4,500 Developer Years, and reduced infra costs by $260M. In a similar manner, AWS states that Amazon Q can also modernise legacy Windows .NET applications by moving to Linux, helping to save on licensing costs and reduce security issues and overheads of patching. One of the more significant parts of the migration announcement was AWS’ pitch to use Amazon Q to transform VMWare and Mainframe applications into cloud-native architectures. Recognising that a one-click solution is not possible with these more complex paradigm shifts, AWS states that Amazon Q can help provide a detailed plan to help teams migrate. These include steps for analysing, documenting and refactoring. While it is still early in the process, AWS aims to shift the timelines for complex migrations, such as mainframe migrations, to be rolled out over several months instead of several years.

As a developer who has worked across numerous projects and organisations, I’ve encountered countless applications stuck on outdated technologies. There are many reasons why upgrades - like migrating Java versions - are often sidelined: they’re not seen as critical, other priorities take precedence, or it can seem daunting to make changes to systems that are performing reliably in production. Yet, these migrations are essential - for improved performance, access to modern functionality, and maintaining critical security support.

Having been involved in migrating legacy apps myself, I know how manual and tedious these tasks can be. That’s why I view AWS’s updates on migration tools with such interest. They have the potential to transform my experience as a developer, freeing me up to focus on more complex, impactful, and business-driven challenges.

Conclusion

This year’s announcements demonstrate AWS’s commitment to addressing past challenges, particularly in the AI space, while expanding its offerings across foundational models, data and analytics tools, and migration technologies. I expect these series of innovations to really deliver tangible value for businesses and developers alike.

As for the next steps, I’m eager to explore scenarios with the wider Kainos community, where these new technologies can be applied to both existing and potential use cases. The updates to Nova are particularly exciting, and we’re keen to evaluate how these models stack up against current solutions in the market.

Register interest for our free AI launchpad workshop to kickstart your AI journey with Kainos. Our experts will help you identify and prototype impactful AI use cases in just 5 days.